search

flowersteam / Grounding_LLMs_with_online_RL

We perform functional grounding of LLMs' knowledge in BabyAI-Text
MIT License
221 stars 24 forks source link

About running the experiment #23

Open Ziyu0118 opened 6 months ago

Ziyu0118 commented 6 months ago

Dear Clement,

I am recently trying to run your experiment. First of all, thanks for your great work! I really appreciate it. Yet I have two small questions:

  1. I saw there are several Slurm scripts in the campaign folder. Does it mean that it is only possible to be implemented on the HPC platform?

  2. After you ground LLM to enhance the sample efficiency, is it possible to be applied in a real environment?

Thanks in advance!

ClementRomac commented 6 months ago

Hi,

Thanks for your interest!

We used a Slurm cluster to perform our experiments, but this is not mandatory at all! What do you mean by a "real environment"?

Ziyu0118 commented 6 months ago

Thank you so much for your kind and prompt reply! That is to say, if we gonna run experiments on our personal laptop, then it is just to run experiments/clm_behavioral-cloning.py, correct?

Then "real environment" in the second question, sorry for my confusing expression, what I want to express is the "real world", like have this project implemented on some real robots like UR3? Thanks in advance for your time and patience

ClementRomac commented 5 months ago

Hey,

Yes you're right!

About the application of such methods with real robots, my answer would be on speculation as we've never done it. What matters I guess is how the environment on which you ground the LLM is close to what the LLM will experience in the robot.

Ziyu0118 commented 2 months ago

Thank you for your answer, then I tried implementing your code without using Slurm. I just used the command "python train_language_agent.py". Yet the error is shown as:

image

Since I don't have the access to the workstation, then how can I deal with it?